Count only elective deferrals of self-employed plan contributions in the saver's credit

[MaxGhenis]

Draft: still waiting on the local test runs (queued for the heavy lock), the impact run (queued) and the independent review.

Problem

The saver's credit list gov.irs.credits.retirement_saving.qualified_retirement_savings_contributions counts self_employed_pension_contributions. That is the person's whole Schedule 1 line 16 contribution to their own SEP, SIMPLE or qualified plan.

The credit counts less:

• 26 U.S.C. 25B(d)(1)(B)(ii)(I) counts "any elective deferrals (as defined in section 402(g)(3))".
• 402(g)(3) defines those as 401(k) cash-or-deferred contributions, salary reduction SEP (SARSEP) contributions, 403(b) salary reduction contributions and SIMPLE IRA elective contributions.
• 402A(a)(1) treats designated Roth contributions as elective deferrals.
• Excluded:
  • nonelective SEP and profit-sharing contributions a self-employed person makes for themself;
  • their matching contributions to their own 401(k) (402(g)(8));
  • their matching contributions to their own SIMPLE IRA (408(p)(9)).

Statute text was checked on 2026-10-11 against the axiom-corpus rows us/statute/26/25B/d/1, us/statute/26/402/g/3 and /8, us/statute/26/402A/a/1 and us/statute/26/408/p/9, and against govinfo USCODE-2023 §§ 25B and 402.

The IRS says the same:

• Form 8880 line 2 instructions (2025, p. 2; 2022, p. 2): "Elective deferrals (including designated Roth contributions under section 402A, if applicable) to a 401(k), 403(b), governmental 457(b), SEP, SIMPLE, or the federal TSP".
• Pub. 590-A (2025, p. 47; 2022, p. 45) lists "salary reduction contributions (elective deferrals …)" to a 401(k), 403(b), governmental 457 plan, SIMPLE IRA plan or salary reduction SEP.

So a sole proprietor whose SEP contribution PE counts in full should get no credit for it.

Decision: count line 16 as elective deferrals first, up to the remaining 402(g) room

The modeled input doesn't say which plan the person has, and Microcosm doesn't carry plan type. New self_employed_pension_elective_deferrals:

min(self_employed_pension_contributions,
    max(elective_deferral_limit − traditional/Roth 401(k) and 403(b) deferrals, 0))

The saver's credit list now counts this instead of self_employed_pension_contributions. elective_deferral_limit is the 402(g)(1) amount plus the person's 414(v) catch-up amount.

Why deferrals first:

• It is the most of line 16 the law could let be elective deferrals. 402(g)(1) caps a person's deferrals across all plans, so whatever exceeds the room their employer-plan deferrals leave must be employer (nonelective or matching) money.
• It assumes the same plan as the contribution limit. Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121 computes the limit assuming a profit-sharing plan with a 401(k) arrangement into which the person defers all remaining room. Only deferrals can take a contribution above the SEP and profit-sharing limit, so any other split would contradict that limit.
• It never cuts a credit the law could allow for some plan the person could have.

Alternatives considered:

• Count none (assume a SEP). This cuts credits for solo 401(k) and SIMPLE savers, whose deferrals do count, and contradicts Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121's limit for contributions above 20% of earned income.
• Count only what can't be a SEP contribution: line 16 less min(415(c) dollar limit, 20% of earned income). This is the lower bound. It depends on Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121's earned income and assumes employer contributions come first, which is not how a solo 401(k) is usually funded.
• An explicit elective-deferral input. No dataset would populate it. A user who knows their plan can already set self_employed_pension_elective_deferrals directly, for example 0 for a SEP.

The variable's documentation says all of this.

What it changes in practice. Take one person with other-plan deferrals W, room R = limit − W, line 16 contribution S, IRA contributions I, ABLE contributions A and testing-period distributions D. The fix changes the credit base only when S > R. The base then falls from max(I + W + S + A − D, 0) to max(I + limit + A − D, 0). The credit caps each person's base at $2,000, so the credit itself changes only when D > limit + I + A − $2,000. That is at least $21,500 in 2025 and $22,500 in 2026, more with catch-up. For a credit-eligible saver, the fix is mostly a correctness fix to the base and rarely moves the credit; the YAML and impact sections show both.

Change

• New policyengine_us/variables/household/expense/retirement/self_employed_pension_elective_deferrals.py (Person, YEAR).
• qualified_retirement_savings_contributions.yaml: self_employed_pension_contributions → self_employed_pension_elective_deferrals in the 2018 and 2026 lists. Also: description typo ("IRA" → "IRS"); references now cite 402(g)(3), Pub. 590-A 2025/2022 and Form 8880 2025/2022 at stable irs-prior URLs. The old references were the current-year PDFs with stale page numbers.

Invariants

policyengine_us/tests/unit/test_savers_credit_self_employed_deferral_properties.py tests these with Hypothesis. Each draw is a batch of 4–15 single filers over 2022, 2025 and 2026, run as one vectorized simulation. The reference uses the IRS-published limits (Notices 2021-61, 2024-80, 2025-67; 414(v)(2)(E) for ages 60–63), not model parameters. It takes from the model only the IRA contributions and the line 16 contribution after its own limit, neither of which this PR changes.

1. counted = min(line 16, room); 0 ≤ counted ≤ line 16; counted ≤ room. Employer-plan deferrals equal min(desired, limit).
2. Employer-plan deferrals + counted ≤ the 402(g) limit plus catch-up.
3. base = max(IRA + employer-plan deferrals + counted − distributions, 0). It is never above the base that counts all of line 16, and falls short of it by at most line 16 − counted.
4. 0 ≤ the person's credit ≤ 50% × min(base, $2,000).

A deterministic grid (age 55, five employer-plan deferral levels × five line 16 amounts) checks that counted rises with line 16, falls with employer-plan deferrals, and is 0 once those deferrals fill the room.

Tests

• self_employed_pension_elective_deferrals.yaml (new, 9 cases, hand-computed from the published limits):
  • no other deferrals;
  • room used up;
  • partial room;
  • catch-up at 55, and the higher catch-up at 61;
  • Roth 403(b) + traditional 401(k) sharing the room (2026);
  • desired deferrals above the limit;
  • 2022;
  • no contributions.
• savers_credit_qualified_contributions.yaml:
  • The 2026 case ($100,000 of profit, $100,000 desired) now expects 24,500, not 72,000. Its self_employed_pension_contributions: 72_000 assertion is dropped: it tests the plan limit, which has its own tests and which Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121 changes.
  • 4 new cases: room used up (23,500), partial room with IRA (25,500), catch-up (31,000), and distributions above the deferral base (0).
• savers_credit_person.yaml: a 2025 single filer at 50%, with 401(k) deferrals of 23,500, line 16 of 5,000 and distributions of 22,000. Base 1,500, credit 750; main gives 1,000.
• Fails on main (main's source with these tests, run from a snapshot; logs in ~/reviews/savers-credit-se-deferrals-2026-10/fails-on-main/): pending (queued for the heavy lock).
• With Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121: every new expectation was computed on a snapshot of Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121 merged with this branch (git merge-tree, clean) and holds. Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121 now edits two cases of savers_credit_qualified_contributions.yaml (its head f59fbe5). When they conflict, take this PR's version of the 2026 case and Limit self-employed plan contributions to the Pub. 560 maximum deduction #10121's case 1.
• Local runs: pending (each file queued separately under heavy_run.sh).

Readers checked

qualified_retirement_savings_contributions has one reader, savers_credit_qualified_contributions. The other readers of self_employed_pension_contributions don't change:

• the ALD;
• ira_active_participant;
• the MN homestead and UT homeowner/renter household income lists, which count all of line 16 as income adjustments, not as saver's credit contributions;
• the state AGI lists that use the ALD.

Impact

Pending: a base-vs-PR microsimulation pair for 2026 is queued for the heavy lock (~/reviews/savers-credit-se-deferrals-2026-10/impact/). Microcosm's SE pension inputs are about 3% of SOI dollars (PolicyEngine/microcosm#1117 item B1), and the arithmetic above limits where the credit can move, so expect a very small change.

Follow-ups

• ABLE contributions (chipped). From 2026 the same list drops able_contributions_person. But the amended 25B(d)(1)(A) (corpus row us/statute/26/25B/d/1) and the 2025 Form 8880 "What's New" (OBBBA) keep ABLE contributions in every year, and from 2027 only ABLE contributions count (SECURE 2.0 § 103; Replace Savers Credit with Savers Match effective 2027 #5199).

axiom: TheAxiomFoundation/rulespec-us#1680 queued. us/statutes/26/25B.yaml takes elective_deferrals as a free input and has no 402(g) module, and no companion test exercises the input while the window is open.

🤖 Generated with Claude Code